SaaS companies are realizing they're building for the wrong user—the human was just a placeholder for the agent that's coming next.

The Summary

  • Lovable, the AI-powered web app builder, is pivoting to "capabilities"—MCP-powered tools that let AI agents actually use software, not just generate it
  • The shift signals a broader reckoning: if agents are doing the work, they need APIs designed for machine-to-machine interaction, not human GUIs
  • This isn't about making better chatbots. It's about rebuilding software infrastructure so agents can act, not just advise

The Signal

Lovable made its name building web apps from natural language prompts. You describe what you want, the AI codes it, you ship. Simple. But CTO Fabian Hedin sees the next move: if agents are building apps, they'll also need to use apps. And most SaaS tools today are designed for humans clicking buttons, not machines making API calls at scale.

The company is now building "capabilities" on top of the Model Context Protocol (MCP). Think of MCP as a standard interface layer that lets AI agents interact with any software service without custom integrations for each one. Instead of a human logging into Stripe to process a refund, an agent calls a Stripe capability and handles it directly.

"The future of SaaS is apps that agents can use."

This matters because the current SaaS model is a mismatch. Software companies spent 20 years optimizing for human workflows: dashboards, drag-and-drop interfaces, notification badges. But if your customer is an AI agent running 10,000 tasks overnight, none of that matters. The agent doesn't care about your UI. It needs:

  • Structured data in, structured data out
  • Predictable error handling
  • Rate limits that match machine speed, not human patience

Lovable's pivot isn't unique. It's a canary in the coal mine. Every SaaS company will face this question: are we building for humans who occasionally automate, or for agents that occasionally need human approval?

Key inflection points:

  • Companies that gate features behind GUIs will lose to those with agent-first APIs
  • MCP adoption could standardize how agents access third-party services, the same way REST APIs standardized web services
  • The "seat-based pricing" model dies when your customer is an agent that never sleeps

The infrastructure shift is already happening. Stripe has an API-first culture. Twilio does. But most SaaS companies bolted APIs onto human-centric products as an afterthought. That won't cut it when the primary user is non-human.

The Implication

If you're building SaaS, audit your product for agent-readiness now. Can an AI agent authenticate, execute core workflows, and handle errors without a human babysitting it? If not, you're building for a user base that's shrinking. The companies that win the next decade will be the ones that treat agents as first-class users, not afterthoughts.

For developers, this is the moment to learn MCP and similar protocols. The job isn't coding apps anymore. It's coding capabilities that agents can compose into apps.

Sources

Latent Space